Skip to content

Graphitization of amorphous carbons: A comparative study of interatomic potentials

Evidence and attribution

Authority of statements

Prose sections below (Summary, Methods, Findings, etc.) are curated summaries of the publication identified by doi, title, and pdf_path in the front matter above. They are not new primary claims by this wiki.

For definitive numerical values, reaction schemes, and interpretations, use the peer-reviewed article (and optional records under normalized/papers/ when present)—not this page alone.

Summary

A fair, single-code comparison of six widely used carbon potentials—Tersoff, REBO-II, ReaxFF, EDIP, LCBOP-I, and COMB3—all run via LAMMPS. The authors note that many carbon potentials exist in the literature but usability in community MD software matters in practice; here every model is exercised as implemented in LAMMPS without author-side modifications, and the long-range AIREBO extension is mentioned as unavailable during their debugging window, which excludes that family from the six-way benchmark. Liquid quenching generates amorphous carbon at several densities, providing a stringent test of whether each potential reproduces the experimental relation between density and hybridization content; subsequent high-temperature annealing probes graphitization and tests each potential’s transferability for bond making/breaking and structural evolution toward layered sp²-rich motifs.

Methods

Comparative classical / reactive MD in LAMMPS benchmarks published carbon models—not a new reactive parametrization.

1 — MD application (shared protocol). Engine: LAMMPS for Tersoff, REBO-II, ReaxFF, EDIP, LCBOP-I, and COMB3 as shipped (no author-side forks). AIREBO: N/A — excluded because LAMMPS AIREBO routines were buggy / unusable during the authors’ tests. System: 32,768 carbon atoms on a slightly randomized simple-cubic lattice at each target density (1.5, 2.0, 2.5, 3.0 g/cm\(^3\)). Boundaries: three-dimensional PBC on the bulk supercells described in Carbon 109 §3. Liquid creation: randomized lattice collapses and melts; Tersoff / REBO / EDIP follow NVE for the first 0.25 ps then apply a thermostat to reach \(T_\mathrm{liquid}\), whereas ReaxFF / LCBOP-I / COMB3 thermostats are engaged from the start to avoid runaway heating. Liquid \(T_\mathrm{liquid}\): 6000 K for REBO-II, EDIP, ReaxFF, COMB3; 8000 K for Tersoff and LCBOP-I; liquid hold 5 ps (15 ps for LCBOP-I). Quench to a-C: 1 ps cool to 300 K, 4 ps equilibration at 300 K before analysis. Annealing: up to 400 ps at case-specific \(T_\mathrm{anneal}\) chosen from MSD diagnostics (tabulated with melt/anneal temperatures in the paper’s Table 2). Ensembles: NVT for the density-series amorphous/graphitization legs described above. Timesteps: 0.05 fs during liquid creation (0.01 fs for LCBOP-I); 0.2 fs thereafter except ReaxFF uses 0.1 fs for annealing. Barostat / pressure: N/A — cells are prepared at fixed target densities in NVT, not stress-controlled NPT. Electric field: N/A — not used. Enhanced sampling: N/A — conventional MD only.

2 — Force-field training. N/A — published parameter sets are evaluated, not refit here.

3 — Static QM. N/A — not used.

4 — Post-MD analysis. Conjugate-gradient energy minimization precedes structural metrics; coordination (1.85 Å cutoff), Franzblau ring statistics, RDFs, and Debye \(I(s)\) (treating the supercell as an isolated cluster for diffraction) follow §3 of the article.

Findings

Outcomes. After the shared melt–quench–anneal pipeline, the six potentials show a wide spread in sp\(^2\) fraction, RDF shape, morphology, ring statistics, and 002-related graphitization metrics. No single model is uniformly best; some are particularly poor for these tests.

Comparisons. Amorphous sp\(^3\) vs density diverges from cited experiment most strongly at high density (Figure 2; Table 3 lists coordination fractions at a-C and at 200 / 400 ps anneal checkpoints). COMB3 low-density a-C shows excess triangular motifs compared with the other five potentials (Figure 3–4 narrative).

Sensitivity / levers. Observables depend on target density and on \(T_\mathrm{liquid}\) / \(T_\mathrm{anneal}\) choices, which must differ among potentials because effective melting points differ.

Limitations / outlook (authored). The authors relate poor transferability to functional-form details and outline needs for next-generation carbon potentials.

Corpus honesty. This page does not duplicate Table 2–3 numerics; see the PDF for exact percentages and diffraction curves.

Limitations

  • “Fair comparison” still depends on parameter files and LAMMPS variant details for each potential family.
  • Reactive potentials may need different timestep/temperature practical limits; the paper should be consulted for run parameters.

Relevance to group

Benchmark-style study placing ReaxFF alongside other carbon reactive/classical models—useful when judging when ReaxFF is appropriate for amorphous carbon and graphitization scenarios.

Citations and evidence anchors

Reader notes (navigation)